A Safety Helmet Detection Model Based on YOLOv8-ADSC in Complex Working Environments

A safety helmet is indispensable personal protective equipment in high-risk working environments. Factors such as dense personnel, varying lighting conditions, occlusions, and different head postures can reduce the precision of traditional methods for detecting safety helmets. This paper proposes an improved YOLOv8n safety helmet detection model, YOLOv8-ADSC, to enhance the performance of safety helmet detection in complex working environments. In this model, firstly, Adaptive Spatial Feature Fusion (ASFF) and Deformable Convolutional Network version 2 (DCNv2) are used to enhance the detection head, enabling the network to more effectively capture multi-scale information of the target; secondly, a new detection layer for small targets is incorporated to enhance sensitivity to smaller targets; and finally, the Upsample module is replaced with the lightweight up-sampling module Content-Aware ReAssembly of Features (CARAFE), which increases the perception range, reduces information loss caused by up-sampling, and improves the precision and robustness of target detection. The experimental results on the public Safety-Helmet-Wearing-Dataset (SHWD) demonstrate that, in comparison to the original YOLOv8n model, the mAP@0.5 of YOLOv8-ADSC has increased by 2% for all classes, reaching 94.2%, and the mAP@0.5:0.95 has increased by 2.3%, reaching 62.4%. YOLOv8-ADSC can be better suited to safety helmet detection in complex working environments.